Flama

Flama is an open source Python framework for putting machine learning models behind an HTTP API. A trained model is packaged into a portable file, a single command serves it, and the result is a production API with a Rust powered core underneath, so the integration work that usually sits between a notebook and a running service is handled by the framework. It covers both predictive and generative models. Models trained in mainstream frameworks such as TensorFlow and PyTorch are packaged into one portable format, so every model looks the same to the API no matter where it came from, and a single command pulls a model from a hub such as HuggingFace and serializes it straight into that format. Generative models can be served through the dialects clients already speak, including endpoints compatible with OpenAI, Anthropic and Ollama alongside a native protocol, and any model served with the native dialect ships with a chat interface at a reserved path that streams token by token with no frontend code. Tools can also be exposed to agents over the Model Context Protocol. There is no pricing page and nothing is sold. The project is released under the Apache 2.0 license by Vortico and installed as a Python package, so the only cost is the infrastructure a team runs it on. The name stands for Framework for Lightweight Applications, Models and Automation. The repository is active rather than archived, and both the source and the documentation are linked from the home page.